Key facts
| Tool type | Free OpenAI-compatible request builder and snippet generator |
| Endpoints | Chat completions with streaming, JSON mode and function calling |
| Models | 30+ models selectable from the live catalogue |
| Auth | API key generated in the dashboard; free plan needs no card |
| Snippet output | cURL, Python and JavaScript examples for each test call |
| Response inspection | Raw JSON, token usage, finish reason and latency |
| Free plan | 2 free AI models (plugsky-micro, plugsky-lite), no card required |
| Product status | Live |
TL;DR
- Test the exact call in the browser before writing application code.
- Exercise streaming, JSON mode and function calling, not just plain chat.
- Copy the generated snippet and keep the tested parameters verbatim.
- Check finish reason and token usage to catch truncation early.
- One OpenAI-compatible endpoint covers 30+ models, so the same snippet works across tiers.
How it works, step by step
- Create an API key in the Plugsky dashboard; the free plan includes 2 free models and no card.
- Open the OpenAI-compatible API tester and point it at https://api.plugsky.com/v1.
- Pick a model from the catalogue and write your system and user messages.
- Set temperature, max tokens and any response-format or tool definitions you need.
- Send the request and inspect the raw JSON, token usage and finish reason.
- Repeat with streaming enabled, then with a tool call, to confirm your parser handles both shapes.
- Copy the generated cURL, Python or JavaScript snippet into your codebase and run your tests.
Try it yourself
Open the OpenAI-compatible API tester →
What to test beyond a basic chat call
A basic completion proves the key works; it does not prove your integration. Streaming changes how bytes arrive and how you render them, so test partial chunks and the final usage block. JSON mode should be exercised with your real schema, not a toy example, because models fail differently on nested and optional fields. Function calling needs at least one multi-turn tool loop: model requests a tool, your code returns a result, the model continues. If those three paths work, most production surprises are already covered.
Reading the response like an engineer
Inspect the fields your app will actually consume. finish_reason tells you whether the model stopped or hit the token ceiling, which is the most common cause of mysteriously empty output. The usage block separates prompt and completion tokens and feeds directly into the token calculator. Error shapes matter too: a 401, a 429 and a malformed-tool-arguments error should each map to a different branch in your retry logic, so trigger them deliberately while you are still in the browser.
From tested call to shipped code
The generated snippet is a starting point, not a substitute for structure. Move the base URL and model name into configuration, wrap the client so you can swap providers, and keep timeouts and one retry with jitter. Because the endpoint is OpenAI-compatible, the snippet also works with the official SDKs by changing base_url and the model name. Run the same snippet against two model tiers — a small free model and a frontier model — to see the quality and latency difference on your own prompt before committing to a default.
Honest comparison
| Capability | Plugsky tester | Typical playground | Throwaway script |
|---|---|---|---|
| Streaming test | Yes, with raw chunk inspection | Sometimes | You implement it |
| JSON mode test | Yes, with your schema | Rarely | Manual |
| Function calling test | Yes, multi-turn tool loop | Rarely | Manual |
| Snippet generation | cURL, Python, JavaScript | Varies | You write it |
| Model coverage | 30+ models, one endpoint | Vendor models only | Whatever you wire up |
| Usage visibility | Token usage and finish reason in the response | Partial | You log it |
Frequently asked questions
Is the API tester free?
Yes. The tester is a free browser tool; model calls consume your plan's usage, and the free plan includes 2 free models with no card.
Which endpoint does it call?
It targets the OpenAI-compatible chat completions endpoint at https://api.plugsky.com/v1, so anything you verify there works with the official OpenAI SDKs.
Can I test streaming?
Yes. Enable streaming in the tester to see chunks arrive and confirm your client renders partial output correctly.
Does it support function calling?
Yes. Define tools with JSON Schema, send a prompt, and inspect the tool call and the follow-up turn.
Are my API keys stored?
Treat the tester like any API client: use keys you can rotate and never paste production secrets into shared environments.
Can I compare models on the same prompt?
Yes. Keep the prompt fixed, change the model, and compare output, token usage and latency before choosing a default.
What snippet languages are generated?
cURL, Python and JavaScript, covering the call you just tested with the same parameters.
Is the tester the same as the migration checker?
No. The tester verifies individual calls; the migration checker audits a whole codebase for OpenAI compatibility.